# CRO: the method when your traffic is too low for A/B testing

> A site with 3,000 sessions a month cannot run a classic A/B test. The CRO method that still holds: reliable measurement, friction by step, prioritisation, a 30-day plan.

- Original page: https://dotsland.com/en/cro-low-traffic-method/
- Language: en
- Published: 2023-09-14T17:00:16+02:00
- Updated: 2026-10-19T18:15:43+02:00
- Section: Digital Marketing
- Publisher: Dotsland, https://dotsland.com/en/

---

## In brief

- **A conversion point gained costs less than a traffic point bought.** On a site with 3,000 sessions and 1.5% conversion, moving to 1.9% brings in as much as 800 extra visits a month, with no recurring advertising budget.
- **Nothing starts before reliable measurement.** If your conversion rate is 25% wrong, every decision taken afterwards is too.
- **Below 10,000 sessions a month, a classic A/B test never reaches a conclusion.** You test sequentially on clear-cut changes, watch session recordings and run five real users through the site.
- **Prioritisation weighs more than creativity.** Three criteria (impact, effort, confidence) are enough to rank a backlog and avoid months lost on micro-changes.

An online shop shows us its dashboard: 2,800 sessions a month, 1.4% conversion, €1,200 of Google Ads budget. The request always goes the same way: how do we bring more people in. Yet the interesting figure is the other one: out of 2,800 visitors, 2,761 leave without doing anything.

CRO (conversion rate optimisation) means working on those 2,761 people rather than paying for 800 more. The principle fits in one sentence. Applying it in a small business is far less obvious, because almost everything published on the subject assumes massive traffic and a testing platform costing several thousand euros a year.

At a few thousand sessions a month, the method changes: fewer tools, more clear-cut changes, strict measurement discipline. The main constraint is not budget, it is statistics.

## A conversion point is worth more than a traffic point

Take three numbers you already have to hand: monthly sessions, conversion rate, average basket. The same gain comes by two routes whose costs are not remotely comparable.

*Simulation for a site with 3,000 monthly sessions and an €80 average basket. Redo it with your own figures.*

| Route | What changes | Monthly result | Recurring cost |
| --- | --- | --- | --- |
| Starting point | 3,000 sessions, 1.5%, €80 basket | 45 orders, €3,600 | None |
| Conversion (CRO) | Rate raised to 1.9%, i.e. 0.4 points better | 57 orders, €4,560 | None, once delivered |
| Acquisition | 800 more sessions, at €0.80 per click | The same €960 gain | €640 a month, with no end |

The conversion gain repeats every month with no new spending.

An honest nuance: CRO has a ceiling. A general site does not go from 1.5% to 6%, and gains become scarcer as the obvious frictions disappear. The first months are therefore the most profitable.

## The non-negotiable prerequisite: accurate conversion figures

A conversion rate is a ratio between two measurements, and both break easily. A tag installed twice inflates sessions. A purchase event that does not fire on instalment payments wipes out part of the sales.

Three causes come up again and again with our clients: ad blockers that remove the measurement script, consent refusals that cut collection, and a break between the checkout and the confirmation page. The missing conversions are spread unevenly across browsers and devices, which also skews comparisons.

The check takes an hour. Compare the number of orders in your analytics tool with the number in your back office over the last 30 days. A 5% gap is routine; a 25% gap means you are steering blind.

Two checks complete the diagnosis: an [up-to-date tagging plan](https://dotsland.com/en/tagging-plan-web-optimisation/) listing the events expected page by page, and a careful reading of your consent banner, because [badly worded consent makes collection drop](https://dotsland.com/en/cookie-banner-design-and-wording/) far more than the regulation requires. If you work with [Google Analytics 4](https://dotsland.com/en/ga4-badly-configured/), check first that the purchase event reports the amount and the order ID.

When the gap exceeds 20%, browser-side measurement has reached its limits and collection must move server-side. We detailed this mechanism in our article on the [silent lie of browser-side tracking](https://dotsland.com/en/tracking-server-side-lying-in-silence/). To run the diagnosis without commitment, the DataFirefly Server-Side platform, published by Datafirefly Limited, which belongs to the same group as Dotsland, offers a [free plan of up to 10,000 requests a month](https://server-side.datafirefly.com/).

> **The mistake not to make**  
> Starting an optimisation project before checking the measurement. A 25% gap between analytics and back office makes every decision that follows wrong: you will spend weeks on a step that loses nothing while the real leak continues.

## The most frequent frictions, step by step

### Arrival: the page’s purpose is not readable

A page that tries to say everything says nothing. The visitor must understand within a few seconds what you sell, to whom, and what action is expected of them.

What we look at: bounce rate by landing page cross-referenced with traffic source, and consistency between the ad clicked and the title of the landing page. A mismatch in promise between the two produces an immediate bounce that nobody ever attributes to the page.

### Exploration: the visitor cannot find what they are looking for

On a catalogue, navigation and internal search do more damage than design. A search that returns nothing for a product you sell is a lost and invisible sale.

What we look at: internal search terms with no results, the number of pages viewed before exit, and filters never used that clutter the mobile screen.

### Decision: reassurance is missing at the moment of doubt

The blocking questions are always the same: delivery time, shipping costs, return conditions, payment method, whether a human contact exists. If the answer comes after the add-to-basket click, it comes too late.

What we look at: where delivery and returns information sits on the product page, whether a contact is visible, and the reasons customers contact customer service, which reveal what the site failed to explain.

### Action: the form asks for too much

Every field added costs submissions. Mandatory account creation before payment remains one of the best-documented causes of abandonment, and it is often kept for internal reasons rather than for the customer.

What we look at: the number of truly mandatory fields, error messages that appear after submission rather than during entry, and the abandonment rate between the form being displayed and submitted. For baskets abandoned despite everything, an [automated follow-up sequence](https://dotsland.com/en/ecommerce-marketing-automation-3-scenarios/) recovers part of what was lost.

## Testing when you do not have 100,000 visitors

The sample size an A/B test needs can be worked out on the back of an envelope: about 16 × p × (1 − p) divided by the gap squared, where p is your current conversion rate and the gap the absolute increase you want to detect. The result gives the number of sessions per variant, at the usual thresholds of 95% confidence and 80% power.

*Formula above, starting rate 2%, traffic split equally between two variants.*

| Test ambition | Absolute gap | Sessions per variant | Duration at 3,000 sessions a month |
| --- | --- | --- | --- |
| Relative increase of 20% (2% to 2.4%) | 0.004 | about 19,000 | 13 months |
| Relative increase of 50% (2% to 3%) | 0.01 | about 3,100 | 2 months |
| Relative increase of 100% (2% to 4%) | 0.02 | about 770 | 3 weeks |

At thirteen months, your site will have changed several times before the end of the test. The only valid window for a small business is that of clear-cut changes.

Direct consequence: below 10,000 monthly sessions, stop testing a button’s colour. Test removing account creation, going from seven fields to three, showing shipping costs on the product page, a complete redesign of a landing page.

### Three valid methods at low volume

The sequential before/after test compares two equivalent periods around a single change. It requires precautions: same duration, same mix of traffic sources, no promotion in either window, and confirmation over a second cycle. It does not prove causality; it provides enough evidence to decide.

The qualitative user test costs half a day. Five people matching the target profile, one precise task to complete while thinking aloud, no help from you. Five sessions are enough to surface most of the major blockages, an order of magnitude established by Jakob Nielsen’s work on usability.

Analysing session recordings can be done with the free plans on the market. Filter on visitors who reached the basket without ordering, watch twenty sessions in full, note every hesitation. This reading gives no figures; it gives hypotheses, which is exactly what is missing at the start.

> **What to remember**  
> Below 10,000 sessions a month, an A/B test only settles clear-cut changes. Twenty session recordings and five user tests give answers in half a day, where statistics would take a year.

## Prioritising with three columns

A CRO backlog fills quickly and is worked through slowly. Score each idea from 1 to 5 on three criteria: expected impact on revenue, implementation effort, confidence in the hypothesis. The score is calculated as impact × confidence ÷ effort.

Confidence is the criterion everyone forgets. An idea backed by twenty session recordings and three customer service reports scores 5. An idea that came out of a meeting scores 2, even if it seems excellent.

A concrete example: removing mandatory account creation (impact 5, confidence 4, effort 2) scores 10. Redesigning the “About” page (impact 1, confidence 2, effort 4) scores 0.5. The ranking sorts itself out and puts an end to decisions by opinion.

## Four traps that cancel out the work

Concluding too early remains the most costly. Looking at results every day and stopping the test as soon as a variant pulls ahead multiplies false positives: on a test with no real effect, this practice ends up producing an apparent winner in a large majority of cases. Set the duration and volume before launching, then stick to them. The same reflex applies to technical measurement tools, where re-running until you get the right figure is self-deception: we documented it on our own site in [what measurement said about six speed optimisations](https://dotsland.com/en/optimisations-that-changed-nothing/).

Testing micro-changes uses up what little traffic you have for nothing. Moving a button produces an effect of a few per cent, undetectable at your scale, and ties you up for weeks.

Confusing correlation with causation leads to absurd decisions. Visitors who use internal search often convert three times more, but putting search everywhere will not triple anything: those visitors already knew what they wanted to buy when they arrived.

Forgetting mobile skews everything else. A site where 70% of traffic is mobile but whose tests are observed on desktop misses the essentials. Systematically segment your results by device type; conclusions often diverge between the two.

## A 30-day plan you can apply from Monday

1. **Days 1 to 5: make measurement reliable.** Compare analytics orders and back-office orders over 30 days, list missing events, fix or document the gap. As long as the gap exceeds 25%, no other action makes sense.
2. **Days 6 to 10: build the funnel.** Measure the pass-through between each step (product page, basket, delivery, payment, confirmation) and identify the two highest steps, by device.
3. **Days 11 to 15: look for causes.** Twenty recordings of abandoned sessions on the two steps identified, five thirty-minute user tests, a reread of the last ten messages to customer service.
4. **Days 16 to 18: rank.** Score all hypotheses on impact, effort and confidence, keep the top three and write down for each the expected result, with a figure, before touching the site.
5. **Days 19 to 30: deliver and measure.** Roll out the best-ranked change, leave it running for two full weeks without touching it, compare with the previous equivalent period and document the result, even if it is negative.

This cycle repeats every month. Its value does not come from any single month but from accumulation: twelve documented clear-cut changes are worth more than an annual redesign whose effect nobody knows.

## Frequently asked questions

### What is a good conversion rate for an online shop?

Published ranges vary too much with sector, traffic mix and average basket to serve as a reliable benchmark. The only useful comparison is your own site, month after month, segmented by traffic source and device. A site at 0.8% whose traffic comes 80% from awareness campaigns may be better managed than a site at 3% that lives only on its brand.

### How many visitors do you need for a reliable A/B test?

It depends on the size of the gain you are aiming for, not on a universal threshold. To detect a relative increase of 20% on a base rate of 2%, allow about 19,000 sessions per variant. To detect a doubling of the rate, about 770 sessions per variant are enough. Below 10,000 monthly sessions, keep A/B tests for structural changes and use qualitative methods for the rest.

### Can you do CRO without a paid tool?

Yes, for the first six months. A correctly configured analytics tool, free session recording plans, a one-question survey on exit pages and five user tests per quarter cover most of the diagnosis. Paid testing platforms become useful when traffic allows several experiments to run in parallel, which assumes several tens of thousands of sessions a month.

### Does CRO work for a B2B site with few forms filled in?

Yes, but the measurement changes focus. With ten to thirty enquiries a month, no statistical test is possible on final conversion. You then work on intermediate indicators (downloading a document, viewing the pricing page, time spent on the method page) and above all on the quality of the enquiries received, which is judged with the sales team rather than in a dashboard.

### How long before you see the first CRO results?

The first 30-day cycle produces a diagnosis and a delivered change, rarely a measurable and certain gain. The effects become readable after three to four cycles, when several clear-cut changes add up on the same funnel. The projects that fail are almost always the ones that stop after the first month, for lack of an immediate result.

## Where to start in your business

The right first question is not “which page should we redo” but “at which step do we lose the most people, and why”. A correctly measured funnel answers the first half within a few days. Twenty session recordings and five users answer the second.

This work belongs to our [UX and CRO expertise](https://dotsland.com/en/expertise/ux-and-cro/), where we combine it with measurement reliability, without which no decision holds. The sequence is always the same: measure accurately, understand where it gets stuck, change things decisively, document.

**Unsure how reliable your conversion rate is, or which step costs you the most?** In a first meeting, we compare your analytics orders with your back office, rebuild your funnel by device and tell you which two steps deserve your next month of work. [Let’s talk](https://dotsland.com/en/contact-us/).